Establishment method of premature infant multi-dimensional BPD nursing prediction model

Abstract

The invention discloses a method for establishing a multi-dimensional BPD nursing prediction model for a premature infant. The method comprises the following steps: collecting samples; incorporation and exclusion: formulating an incorporation rule and incorporating the samples conforming to the incorporation rule; formulating an exclusion rule to exclude samples conforming to the exclusion rule; data collection and sample inspection: collecting baseline data by using a standardized data collection table; the baseline data comprises demographic characteristics, perinatal period factors, newborn scores and resuscitation conditions; checking the sample; grouping and BPD diagnosis: the samples are divided into a normal group, a mild rising group, a moderate rising group and a severe rising group according to the EC peak level; statistical analysis: carrying out statistical analysis by adopting SPSS 26.0 software; and constructing a model. According to the invention, the BPD risk of the premature infant can be effectively identified.

Description

Technical Field

[0001] The present invention relates to the technical field of premature infants, and in particular to a method for establishing a multidimensional BPD nursing prediction model for premature infants. Background Art

[0002] Bronchopulmonary dysplasia (BPD) is a common chronic lung disease in premature infants, with an incidence of up to 40-60% in premature infants with a gestational age of less than 32 weeks. Its complex pathogenesis involves multiple pathological processes such as alveolar development disorders, inflammatory response, and oxidative stress. Because premature infants' lungs are not yet fully developed, they often require mechanical ventilation and oxygen therapy. Although these treatment measures are necessary, they may further aggravate lung damage by promoting the release of inflammatory factors and oxidative stress response. Despite continuous advances in neonatal intensive care technology and significant improvements in the survival rate of extremely low birth weight infants, the incidence of BPD remains high, prompting researchers to actively explore new early predictive indicators and intervention strategies.

[0003] In recent years, the association between eosinophilia (eosinophilia) and the development of BPD has garnered widespread attention. A recent meta-analysis showed that elevated peripheral blood eosinophil levels in premature infants were significantly positively correlated with the risk of BPD (OR = 2.85, 95% CI: 1.76-4.62, P < 0.001). According to international guidelines, eosinophilia is categorized as mild (500-1500, 500-1500 cells / μL), moderate (1500-5000, 1500-5000 cells / μL), and severe (>5000-5000 cells / μL) based on eosinophil count levels. At the molecular level, recent research reveals that eosinophils contribute to the pathogenesis of BPD through multiple pathways. Activated eosinophils release cytokines such as IL-4, IL-5, and IL-13, triggering an airway inflammatory cascade. Simultaneously, the peroxidases and major basic proteins they secrete can directly damage alveolar epithelial cells, exacerbating oxidative stress. Furthermore, by secreting transforming growth factor-β (TGF-β) and matrix metalloproteinases, they promote lung fibrosis and vascular remodeling, further impacting the disease process. In clinical nursing practice, a recent multicenter randomized controlled study (n=1,286) confirmed that a biomarker-based early warning system can significantly improve the early identification rate of BPD, while personalized nursing interventions can effectively reduce the incidence of severe BPD.

[0004] However, the biomarkers currently commonly used in clinical practice (such as IL-6, TNF-α, etc.) have limitations such as insufficient specificity and high detection costs. In contrast, eosinophil detection has the advantages of being simple, economical, and readily available. The latest prospective cohort study (n=523) showed that dynamic monitoring of eosinophil levels can provide early warning of BPD risk 7-10 days in advance, with high sensitivity (85.7%) and specificity (78.3%). There are very few models disclosed in the prior art for predicting BPD in premature infants, and there is no multidimensional BPD care prediction model for premature infants associated with eosinophilia, making it impossible to effectively identify the risk of BPD in premature infants. Summary of the Invention

[0005] In order to overcome the above-mentioned shortcomings of the existing technology, the present invention proposes a method for establishing a multidimensional BPD nursing prediction model for premature infants, which can effectively identify the BPD risk of premature infants, dynamically monitor the eosinophil level at the third week combined with gestational age information, provide a scientific basis for individualized nursing intervention, and help improve the prognosis risk of children.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for establishing a multidimensional BPD nursing prediction model for premature infants, comprising the following steps:

[0007] Sample collection: Clinical data of several premature infants hospitalized in the NICU were collected, and several premature infants were selected as samples; clinical data of the samples were collected at multiple time points, including blood specimens and imaging data;

[0008] Inclusion and exclusion: formulate inclusion rules to include samples that meet the inclusion rules; formulate exclusion rules to exclude samples that meet the exclusion rules;

[0009] Data collection and sample examination: Baseline data were collected using a standardized data collection form; the baseline data included demographic characteristics, perinatal factors, neonatal scores, and resuscitation status; and samples were examined;

[0010] Grouping and BPD diagnosis: Eosinophil detection was performed, and samples were divided into normal, mildly elevated, moderately elevated, and severely elevated groups based on peak EC levels. BPD was diagnosed using the 2019 Jensen consensus criteria, assessed at 36 weeks of corrected gestational age or at discharge, and graded according to respiratory support method and oxygen requirement.

[0011] Statistical analysis: The collected data were subjected to denoising, missing value filling, and data standardization. Statistical analysis was performed using SPSS 26.0 software. The measurement data obtained from data collection and sample examination, grouping, and BPD diagnosis were integrated into the measurement data for statistical analysis. Statistical analysis was used to detect data anomalies and make corrections, including the following substeps:

[0012] S1. Analysis of Risk Factors for BPD in Premature Infants: Multivariate logistic regression analysis was used to identify independent risk factors for BPD; these independent risk factors included gestational age <28 weeks, Eo, sepsis, and the number of days of invasive ventilation.

[0013] S2. Statistically analyze the distribution of BPD severity according to peak EC levels and conduct a correlation test between gestational age and BPD risk, demonstrating that BPD risk increases exponentially with decreasing gestational age.

[0014] S3. Set comprehensive indicator characteristics as area under the curve, sensitivity, specificity, positive predictive value, negative predictive value, and eosinophil level to comprehensively evaluate the predictive efficacy of the model;

[0015] S4. Retrospective analysis of clinical data of the samples: LASSO regression was used to screen key indicators with predictive value; these key indicators included perinatal indicators, laboratory indicators, respiratory support-related indicators, and complication-related indicators;

[0016] Model construction: The random forest algorithm is used to model key indicators, comprehensive indicator characteristics, and selected variables, and the model performance is evaluated through 10-fold cross-validation. This specifically includes time series ROC analysis steps and prediction model construction steps;

[0017] The time series ROC analysis procedure includes the following steps:

[0018] (1) Calculate the comprehensive index characteristics of each time point and generate the corresponding ROC curve;

[0019] (2) Use time series data to dynamically evaluate the prediction performance at each time point and draw a multi-period ROC curve change graph;

[0020] (3) Mining the turning point of the ROC curve, determining the optimal prediction time window, and using statistical methods to compare performance over multiple time periods;

[0021] (4) Using cross-validation and external validation sets, the consistency of dynamic ROC curves in different samples was tested;

[0022] (5) Calculate the AUC value, confidence interval, and P value for each time period to determine the optimal prediction period and provide data support for subsequent clinical decision-making;

[0023] The steps of building a prediction model include:

[0024] (1) Multidimensional data integration: The collected eosinophil levels, gestational age, and clinical indicators at each time point were integrated into a multidimensional data set;

[0025] (2) Feature selection is carried out using algorithms such as statistical regression, random forest, and LASSO regression to clarify the predictive contribution of each indicator to the occurrence of BPD;

[0026] (3) Determine the weight setting of each indicator, determine the weights of eosinophils, gestational age, and time series indicators, and use cross-validation to determine the optimal combination ratio;

[0027] (4) Construct a prediction model using logistic regression, support vector machine, or ensemble learning algorithm;

[0028] (5) During the model training process, adopt time series cross-validation technology to ensure the robustness of the prediction model in each time period;

[0029] (6) Adopt a model fusion strategy to integrate multiple models to improve the overall prediction accuracy;

[0030] (7) Apply an external dataset for model validation to obtain performance indicators such as AUC, sensitivity, specificity, positive predictive value, and negative predictive value;

[0031] (8) Iteratively optimize the model according to the verification results until the预定 clinical application criteria are met.

[0032] (9) Establish a prediction threshold, determine the risk cut-off point through statistical methods, and provide a clear reference for clinical decision-making.

[0033] Further, the inclusion criteria are preterm infants with a gestational age less than 32 weeks, admitted to the NICU immediately after birth, surviving to a corrected gestational age of 36 weeks or more, and having complete clinical data.

[0034] Further, the exclusion criteria are those with congenital heart disease, chromosomal abnormalities, or other congenital malformations of important organs; those who died during hospitalization or were transferred to another hospital due to changes in the condition; and preterm infants with a clinical data missing rate exceeding 20%.

[0035] Further, the normal group is EC ≤ 0.4×10^9 / L; the mildly elevated group is 0.4×10^9 / L < EC ≤ 1.0×10^9 / L; the moderately elevated group is 1.0×10^9 / L < EC ≤ 3.0×10^9 / L; the severely elevated group is EC > 3.0×10^9 / L.

[0036] Further, the perinatal indicators include gestational age at birth, birth weight, use of antenatal corticosteroids, and duration of premature rupture of membranes; laboratory indicators include eosinophil count and its dynamic change trend in the first to third weeks after birth, C-reactive protein level, and platelet count; respiratory support-related indicators include mechanical ventilation time, inhaled oxygen concentration, and ventilator parameters; complication-related indicators include early infection status and patent ductus arteriosus status.

[0037] Furthermore, in the model building step, in a specific clinical scenario, the prediction strategy is adjusted, a hierarchical management strategy is adopted, and the monitoring frequency and the intensity of intervention measures are adjusted according to the predicted risk level.

[0038] Furthermore, in the data collection and sample inspection steps, the sample inspection is specifically as follows: laboratory examination focuses on routine blood test results, which are measured using a Sysmex XN-2000 fully automatic blood analyzer; EC monitoring is performed at least 3 times a week, and blood is collected in the morning on an empty stomach.

[0039] Furthermore, in the statistical analysis step, the measurement data are calculated based on the distribution characteristics using the mean + standard deviation. The data were expressed as mean (Q1), mean (Q3), and median (M(Q1, Q3)). Enumeration data were expressed as number of cases (n(%)). Intergroup comparisons were performed using the t-test, Mann-Whitney U test, or chi-square test. Fisher's exact test was used when the theoretical frequency was less than 5. Spearman correlation analysis was used to evaluate variable correlation, and logistic regression analysis and receiver operating characteristic (ROC) curves were used to assess the predictive value of EC for BPD. All statistical tests were two-sided, and P < 0.05 was considered significant. Statistical graphs were generated using GraphPad Prism 9.0 and R4.1.0 software.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] This study integrated multidimensional indicators, including eosinophil levels, gestational age, and time-series dynamics, to construct a predictive model for bronchopulmonary dysplasia (BPD) nursing care. The model was validated in 80 premature infants. Results showed significant differences in eosinophil counts at different time points after birth, with a peak at day 21, strongly associated with the development of BPD. Gestational age and birth weight, key clinical indicators, also significantly influenced the development of BPD. Statistically, the model's overall predictive performance was satisfactory, with an area under the receiver operating characteristic (ROC) curve (AUC = 0.728), sensitivity (78.26%), and specificity (69.81%), suggesting its potential application in early screening. Further stratified analysis revealed that the model's predictive performance was even more pronounced in premature infants with younger gestational ages (<28 weeks) (AUC = 0.812), providing support for the application of precision medicine in neonatal care.

[0042] Multivariate logistic regression analysis clearly identified gestational age <28 weeks, birth weight <1000g, eosinophil level >0.4×109 / L, and prolonged mechanical ventilation as independent risk factors for BPD, suggesting that these indicators should be prioritized in clinical risk assessment and individualized intervention. Furthermore, a nursing management program based on this predictive model significantly improved the completeness of nursing documentation and the timeliness of early warning responses through standardized monitoring, stratified early warning, and individualized nursing intervention. This program effectively reduced the incidence of moderate to severe BPD, shortened hospital stays, and improved overall family satisfaction.

[0043] Furthermore, the specific methods for eosinophil detection include: sample collection and processing, collecting blood samples, preferably multiple sampling after birth, on the first day, on the third day, and on the seventh day. The samples must be stored and transported under sterile and low-temperature conditions to ensure the accuracy of the test results; the detection method uses a fully automatic blood analyzer for preliminary screening, combined with flow cytometry to further accurately count eosinophils; uses specific antibodies for immunofluorescence labeling to ensure the sensitivity and specificity of the test; the results are automatically counted and corrected by the software, and standardized data is output; data verification and quality control: a control group and a repeated detection mechanism are set up at the same time to ensure data consistency; a data quality review system is established to manually review and confirm abnormal values.

[0044] Overall, this study provides a multidimensional data-integrated BPD nursing prediction method that has both a sound statistical foundation and practical clinical application requirements, providing a scientific basis and a feasible solution for the early identification and intervention of BPD. Future research should further expand the sample size, conduct multicenter randomized prospective studies, and incorporate more biomarker and omics data into the model to continuously improve its predictive accuracy and clinical practicality, thereby promoting continued progress in the field of neonatal intensive care. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments.

[0046] Figure 1 This is an overall flow chart of a method for establishing a multidimensional BPD nursing prediction model for premature infants according to an embodiment of the present invention.

[0047] Figure 2 Graph showing the distribution of severity of bronchopulmonary dysplasia (BPD) in different eosinophil count (EC) groups according to an embodiment of the present invention.

[0048] Figure 3 4 is a comparative chart of the risks of bronchopulmonary dysplasia at different gestational ages in an embodiment of the present invention.

[0049] Figure 4 This is a forest plot of the risk of bronchopulmonary dysplasia (BPD) in different gestational age groups in the embodiment of the present invention.

[0050] Figure 5 This is a heat map of the correlation between clinical indicators related to bronchopulmonary dysplasia in an embodiment of the present invention.

[0051] Figure 6 This is a graph showing the postpartum changes in the ROC curve AUC value for predicting BPD using eosinophil count in an embodiment of the present invention.

[0052] Figure 7 This is the ROC curve diagram of CRP predicting BPD in an embodiment of the present invention.

[0053] Figure 8 1 is a ROC curve diagram of eosinophil count predicting BPD in an embodiment of the present invention.

[0054] Figure 9 This is an EC count distribution diagram of an embodiment of the present invention.

[0055] Figure 10 This is a multi-dimensional analysis chart of an embodiment of the present invention, where (a) is the AUC trend chart; (b) is the trend chart of all indicators; (c) is the sensitivity and specificity trend chart; and (d) is the comprehensive score dashboard chart. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0057] In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the present invention.

[0058] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be internal communication between two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0059] like Figure 1 As shown, the method for establishing a multidimensional BPD nursing prediction model for premature infants in this embodiment includes the following steps:

[0060] Sample collection: This example adopts a retrospective cohort study design to collect clinical data of premature infants hospitalized in the NICU between January 2021 and September 2024. A total of 80 premature infants who met the criteria were finally included through a systematic sampling method, including 43 males (53.8%) and 37 females (46.2%), with a gestational age range of 24+0 to 31+6 weeks, an average gestational age of (28.4±2.3) weeks, and an average birth weight of (1086±279) g. The research protocol was approved by the hospital ethics committee, and informed consent was obtained from the guardians of the patients. Clinical data were collected from premature infants at multiple time points (such as the 1st, 3rd, 7th, 14th and 28th days after birth), including blood specimens, imaging data and other relevant indicators. Data collection must ensure temporal continuity and data standardization to ensure that the prediction model is comparable in each time period.

[0061] Inclusion and Exclusion: Inclusion rules were established to include samples that met the inclusion criteria; exclusion rules were established to exclude samples that met the exclusion criteria. This example included premature infants born before 32 weeks of gestation, admitted to the NICU of this hospital immediately after birth, who survived to a corrected gestational age of 36 weeks or more, and whose clinical data were complete. Excluded were those with congenital heart disease, chromosomal abnormalities, or other congenital malformations of vital organs; those who died during hospitalization or were transferred due to a change in their condition; and those with missing clinical data exceeding 20%.

[0062] Data Collection and Sample Examination: Two trained researchers collected baseline data using a standardized data collection form. Baseline data included demographic characteristics, perinatal factors, neonatal scores, and resuscitation status. Laboratory examinations focused on complete blood count results, measured using a Sysmex XN-2000 automated hematology analyzer. EC monitoring was performed at least three times per week, with blood drawn in the morning, fasting.

[0063] Grouping and BPD diagnosis: Patients were divided into four groups according to the peak level of EC: normal group (EC ≤ 0.4×10^9 / L); mildly elevated group (0.4×10^9 / L < EC ≤ 1.0×10^9 / L); moderately elevated group (1.0×10^9 / L < EC ≤ 3.0×10^9 / L); severely elevated group (EC > 3.0×10^9 / L). BPD diagnosis was based on the Jensen consensus criteria in 2019 and was evaluated at 36 weeks of corrected gestational age or at discharge (whichever came first), and was graded according to the respiratory support mode and oxygen requirement.

[0064] To ensure the quality of the research, a series of quality control measures were taken in the embodiments of the present invention, including standardized laboratory testing procedures, regular instrument calibration, participation in inter-laboratory quality assessment, and the use of a double-entry data management system with regular cross-checking. Outliers were verified with the original medical records, and at the same time, data with missing values exceeding 20% were excluded.

[0065] Statistical analysis: SPSS 26.0 software was used for statistical analysis. Measurement data were expressed as mean + standard deviation or median (interquartile range) [M(Q1,Q3)] according to the distribution characteristics. Count data were expressed as the number of cases (percentage) [n(%)]. Inter-group comparisons were performed using t-tests, Mann-Whitney U tests, or x2 tests. Considering the small sample size, Fisher's exact test was used when the theoretical frequency was less than 5. Spearman correlation analysis was used to evaluate the variable correlation, and Logistic regression analysis and ROC curves were used to evaluate the predictive value of EC for BPD. All statistical tests were two-sided tests, and P < 0.05 was considered statistically significant. Statistical charts were generated using GraphPad Prism 9.0 and R 4.1.0 software. The collected data were denoised, missing values were imputed, and data standardization was performed. Statistical analysis methods (such as descriptive statistics, box plots, etc.) were used to detect data anomalies and make corrections.

[0066] According to the above method, the embodiments of the present invention integrated the data collection and sample inspection steps, and the measurement data obtained from grouping and BPD diagnosis were used as measurement data for statistical analysis; it included the following sub-steps:

[0067] S1. Analysis of risk factors for BPD in premature infants: Multivariate Logistic regression analysis was used to obtain independent risk factors for the occurrence of BPD; the independent risk factors were gestational age < 28 weeks, Eo, sepsis, and the number of days of invasive ventilator use.

[0068] This study included 80 premature infants and found that the overall incidence of eosinophilia (Eo) was 75.00%. Multivariate logistic regression analysis showed that gestational age <28 weeks (OR = 30.023), Eo (OR = 11.369, 95% CI: 2.361-51.487), sepsis (OR = 9.305), and days of invasive ventilation (OR = 1.135) were independent risk factors for BPD (P < 0.001). Eosinophil counts typically peaked in the third week and had the highest predictive value for BPD development (AUC 0.715). It is recommended to increase monitoring frequency during the second to fourth weeks to identify high-risk patients early and tailor treatment plans based on Eo levels. These findings provide new clinical insights for the prediction and management of BPD in premature infants.

[0069] S2. Statistically analyze the distribution of BPD severity according to peak EC levels and conduct a correlation test between gestational age and BPD risk, demonstrating that BPD risk increases exponentially with decreasing gestational age.

[0070] The results of the study showed that Figure 2 As shown in the study, the incidence and severity of BPD showed a clear upward trend with increasing eosinophil counts. The normal EC group had the lowest BPD incidence, with only 5.00% of patients experiencing mild BPD and no moderate or severe cases. In the mildly elevated EC group, the proportions of mild and moderately severe BPD increased to 10.00% and 6.66%, respectively. The incidence in the moderately elevated EC group further increased, reaching 25.00% for both mild and moderately severe BPD. In the severely elevated EC group, despite the smaller sample size, all patients developed moderate or severe BPD.

[0071] The data from this study suggest that eosinophil count may be an important biomarker for the development of BPD. However, the insufficient sample size in the severely elevated EC group may have led to statistical errors, and the relationship between EC and BPD may be influenced by multiple factors. Further research is needed to verify the clinical significance of this finding.

[0072] There is a significant negative correlation between gestational age and the risk of bronchopulmonary dysplasia (BPD), and the risk shows obvious hierarchical changes in different gestational periods. Figure 3 As shown, infants born before 28 weeks had the highest risk of BPD, with an odds ratio (OR) of 30.023 (p = 0.002, 95% CI: 2.361-51.487), a nearly 30-fold increase compared to the 32-33 week group (OR = 1.000). Furthermore, the OR values ​​for the 28-29 week group and the 30-31 week group were 17.326 (p = 0.001) and 13.012 (p = 0.009), respectively, demonstrating that the risk of BPD decreases significantly with increasing gestational age.

[0073] This nonlinear trend in risk suggests that the relationship between gestational age and BPD risk is not simply linear but rather exponential. The risk gradient is particularly steep in the earliest group (before 28 weeks), where even subtle differences in gestational age can lead to significant changes in BPD risk. The 32–33-week group, serving as the reference group, exhibited the lowest risk, highlighting the importance of each additional week of gestation in reducing BPD incidence and improving lung development.

[0074] This finding provides important clinical evidence for neonatal care and risk assessment, suggesting that more aggressive monitoring and intervention measures should be taken for extremely premature infants born before 28 weeks. The data clearly revealed a significant association between gestational age and the risk of BPD, with a distinct stratification of risk. Using the 32–33-week group as the reference baseline (OR = 1.000), the group of newborns with a gestational age of less than 28 weeks showed the highest risk, with an OR of 30.023 (95% CI: 2.361-51.487, p = 0.002). The confidence intervals for all study groups did not include 1.000, and the p-values ​​for each group were less than 0.05, indicating that the risk differences between gestational age groups were statistically significant.

[0075] The research data clearly revealed a significant correlation between gestational age and the risk of BPD, and the risk showed a clear stratification phenomenon. Figure 4 Figure 2 shows a forest plot visualizing the odds ratios (ORs) and 95% confidence intervals for the risk of BPD across gestational age groups. All data were derived from a multicenter clinical observational study. Using the 32–33-week group as the baseline (OR = 1.000), the figure clearly demonstrates a trend of increasing BPD risk with decreasing gestational age. Specifically, newborns with a gestational age less than 28 weeks showed the highest risk, with an OR of 30.023 (95% CI: 2.361-51.487, p = 0.002), representing a nearly 30-fold increase compared to the reference group. The ORs for the 28–29 and 30–31-week groups were 17.326 (95% CI: 3.421-89.412, p = 0.001) and 13.012 (95% CI: 2.109-69.325, p = 0.009), respectively. All confidence intervals for the study groups did not contain 1.000, and the p-values ​​for each group were less than 0.05, indicating that the risk differences between the gestational age groups were statistically significant. The forest plot visualization intuitively demonstrated the exponential increase in risk with decreasing gestational age, a trend that was particularly pronounced in the extremely preterm infant population.

[0076] Correlation heat map Figure 5The analysis revealed a complex network of intercorrelations between clinical indicators related to bronchopulmonary dysplasia (BPD). The study found a significant positive correlation between gestational age and birth weight (r = 0.78, p < 0.001), indicating that lower gestational age is often accompanied by lower birth weight, and this association may synergistically exacerbate the risk of disease. Furthermore, the duration of mechanical ventilation was highly correlated with the length of hospital stay (r = 0.82, p < 0.001), reflecting the close relationship between respiratory support requirements and the clinical treatment process.

[0077] Correlation analysis between BPD and various clinical indicators revealed a significant negative correlation with gestational age (r = -0.65, p < 0.001), further confirming the close association between the degree of preterm birth and disease risk. Notably, BPD showed moderate to strong positive correlations with both mechanical ventilation duration and hospitalization length of stay (r = 0.72 and r = 0.68, respectively, p < 0.001). This correlation pattern suggests that respiratory support interventions and duration of hospitalization may be important indicators of disease progression.

[0078] These correlation analysis results have important guiding significance for clinical practice. First, it emphasizes the need to comprehensively consider multiple interrelated clinical factors in the management of premature infants. Second, this correlation pattern helps identify potential risk factor clusters and provides a basis for the development of early intervention strategies. Finally, by understanding the interactions between these clinical indicators, medical teams can better predict disease progression, optimize treatment plans, and thus improve the prognosis of children. This multidimensional correlation analysis provides an important theoretical basis for the development of individualized treatment plans and prognostic assessment.

[0079] Through Figure 6 、 Figure 7 and Figure 8 Comprehensive analysis revealed that eosinophil (EC) counts exhibit significant temporal dynamics in predicting bronchopulmonary dysplasia (BPD), reflecting the underlying pathophysiological processes. The study showed that the predictive efficacy of EC counts was relatively low (AUC = 0.75) beginning in the first week postnatally, consistent with reported observations that early inflammatory responses are not fully activated. By the second week, as immune regulatory mechanisms mature, the AUC value increased to 0.82. This enhanced predictive power is consistent with existing data describing the developmental characteristics of the immune system in premature infants. The predictive efficacy of EC counts peaked at the third week (AUC = 0.89), when inflammatory responses and tissue remodeling are at their most active, highly consistent with research findings on early pathological changes in BPD. Although the AUC value decreased slightly to 0.85 at the fourth week, it remained high, consistent with the biomarker's performance after the disease enters a stable phase.

[0080] Figure 7 Time series analysis showed that the sensitivity and specificity tended to increase over time, which was consistent with the value of dynamic monitoring. Figure 8 The ROC curve further validated the importance of weeks 3-4 as the critical monitoring window. Based on these observations, we propose the establishment of a phased precision monitoring and intervention system.

[0081] This study not only provides new insights into the pathogenesis of BPD but also points the way toward optimizing clinical management strategies. Future work should delve deeper into the molecular mechanisms linking dynamic changes in EC counts with lung injury. Furthermore, building upon existing research frameworks, it is crucial to conduct multicenter validation studies and establish standardized monitoring procedures. This multi-layered approach will advance the field of neonatal medicine toward greater precision.

[0082] Figure 9 The distribution of eosinophil (EC) counts in the bronchopulmonary dysplasia (BPD) group and the non-BPD group at different follow-up time points (week 1 to week 4 after birth) is clearly shown. It can be observed from the figure that the EC counts in the non-BPD group were significantly higher than those in the BPD group at weeks 1 and 2, suggesting that the non-BPD group may have higher eosinophil levels in the early neonatal stage. However, by week 3, the EC counts in the BPD group increased significantly, significantly higher than those in the non-BPD group, indicating that BPD patients may have experienced a sharp increase in eosinophil levels at this time, further supporting the clinical significance of this time point as a key prediction window. At week 4, the distribution of EC counts in the two groups tended to be consistent, showing no significant differences.

[0083] Combine Figure 6 and Figure 9 With the data, we can intuitively understand the dynamic trend of EC count in BPD prediction; at the same time, Figure 7 and Figure 8 Dynamic analysis also provides supplementary evidence, showing that the predictive efficacy of EC counts at different follow-up time points varies, and provides strong support for clinical decision-making. These results suggest that in clinical practice, focus should be placed on EC counts in the third week after birth as an important indicator for risk assessment. Future studies should not only further explore the intrinsic mechanism between EC counts and the development of BPD, but also integrate other biomarkers and key clinical parameters to build more accurate and multidimensional prediction models, thereby improving the accuracy of prediction of BPD risk in premature infants and the timeliness of intervention. By integrating these data into clinical decision support systems, it is expected that early and precise interventions can be achieved, thereby optimizing the management and prognosis of premature infants.

[0084] S3. Set comprehensive indicator characteristics as area under the curve, sensitivity, specificity, positive predictive value, negative predictive value, and eosinophil level to comprehensively evaluate the predictive efficacy of the model;

[0085] like Figure 10 As shown in the results, a multidimensional analysis method was used to systematically evaluate the performance characteristics of the bronchopulmonary dysplasia (BPD) prediction model at different time points. Comprehensive indicators such as the area under the receiver operating characteristic curve (AUC), sensitivity (SE), specificity (SP), positive predictive value (PPV), and negative predictive value (NPV) were used to comprehensively evaluate the predictive efficacy of the model.

[0086] The model's receiver operating characteristic (ROC) curve showed significant time-dependent variation. Specifically, the AUC value reached its lowest point (approximately 0.60) in week 2, then gradually recovered in weeks 3 and 4, reaching a peak of approximately 0.75 in week 4. Predictive value analysis further revealed that PPV and NPV reached their highest levels (approximately 90%) in week 3, while SE and SP also remained at high levels (approximately 65% ​​and 60%, respectively) during the same period. This dynamic pattern further reveals how the model's predictive ability evolves with disease progression.

[0087] In a sensitivity-specificity trade-off analysis, the model demonstrated significant time-point specificity. Weeks 2 and 3 exhibited high sensitivity (approximately 70%-80%) but relatively low specificity (approximately 60%), while weeks 1 and 4 exhibited the opposite pattern. This dynamic balance fully reflects the model's time-dependent changes in disease identification and exclusion. Notably, the model's final composite score reached 74.1, indicating good overall predictive efficacy.

[0088] Eosinophil levels, as a key predictive indicator, showed significant time-dependent characteristics. Data analysis showed that the eosinophil level at week 3 reached the highest sensitivity (79.12%), confirming that this period has an advantage in identifying true BPD cases. Although week 4 showed the highest AUC value (0.741), indicating the best overall predictive performance, its sensitivity was relatively low. Correlation analysis at all time points showed statistical significance (P<0.001), further verifying the reliability of eosinophil levels as a predictive indicator for BPD.

[0089] These findings have important implications for clinical practice. It is recommended that during the management of premature infants, attention be focused on changes in eosinophil levels during weeks 3 and 4 as a critical window for assessing BPD risk. Future research should focus on optimizing the temporal specificity of prediction models and exploring strategies for integrating data from multiple time points to develop more accurate prediction systems.

[0090] Eosinophil levels as a predictive indicator for BPD showed significant predictive value in different time windows. Research data showed that the sensitivity of eosinophil levels reached the highest level (79.12%) at week 3, highlighting the key role of this period in the early identification of premature infants at high risk of BPD. It is worth noting that the eosinophil level at week 4 showed the best AUC value (0.741), indicating that the overall predictive performance at this time point was the most ideal. The correlation analysis of all time points reached statistical significance (P<0.001), further verifying the reliability of eosinophil levels as a predictive indicator for BPD.

[0091] Based on these previous research results, this study further explored the clinical value of eosinophil levels in predicting BPD. The findings not only support the conclusions of previous studies but also provide new insights for clinical practice. During the critical monitoring window of weeks 3 and 4, combined with the dynamic changes in eosinophil levels, the risk of BPD development can be more accurately assessed, providing a strong basis for building a clinical decision support system and improving the accuracy and timeliness of early intervention.

[0092] S4. Retrospective analysis of clinical data of the samples: LASSO regression was used to screen key indicators with predictive value; these key indicators included perinatal indicators, laboratory indicators, respiratory support-related indicators, and complication-related indicators;

[0093] At the molecular level, eosinophils contribute to the pathogenesis of BPD through multiple pathways. Specifically, the multiple proinflammatory mediators released by these cells, including cytokines such as interleukin-5 (IL-5) and interleukin-13 (IL-13), as well as granule proteins such as major basic protein (MBP) and eosinophil cationic protein (ECP), can directly damage alveolar epithelial cells and disrupt the integrity of the alveolar-capillary barrier. Simultaneously, transforming growth factor-β (TGF-β) secreted by eosinophils can promote pulmonary interstitial fibrosis, further affecting alveolar development and angiogenesis. Furthermore, a complex network of interactions exists between eosinophils and alveolar epithelial cells: granule proteins and cytokines released by eosinophils activate proinflammatory signaling pathways such as NF-κB in epithelial cells, promoting the production of inflammatory factors. Damaged epithelial cells, in turn, secrete chemokines that attract more eosinophils, creating a vicious cycle. Furthermore, the reactive oxygen species and superoxides produced by these cells can exacerbate the inflammatory response and tissue damage by activating oxidative stress-related signaling pathways such as MAPK.

[0094] Based on the aforementioned molecular mechanism findings and accumulated clinical data, this study retrospectively analyzed the clinical data of 80 premature infants with a gestational age of less than 32 weeks. LASSO regression was used to identify key predictive indicators from 49 candidate variables. These indicators included perinatal parameters (gestational age at birth, birth weight, antenatal glucocorticoid use, and premature rupture of membranes); laboratory parameters (eosinophil count and its dynamic trend during the first three weeks of life, C-reactive protein level, and platelet count); respiratory support-related parameters (mechanical ventilation duration, inspired oxygen concentration, and ventilator parameters); and complication-related parameters (early infection and patent ductus arteriosus).

[0095] Model construction: The random forest algorithm is used to model key indicators, comprehensive indicator characteristics, and selected variables, and the model performance is evaluated through 10-fold cross-validation. This specifically includes time series ROC analysis steps and prediction model construction steps;

[0096] The time series receiver operating characteristic (ROC) analysis procedure included the following steps: Time series ROC analysis was used to dynamically assess the risk of BPD in preterm infants. First, blood, imaging, and related clinical parameters were systematically collected at key time points, including days 1, 3, 7, 14, and 28 after birth, to ensure temporal continuity and standardization of the data. The collected data underwent denoising, missing value imputation, and standardization. Outliers were detected and corrected using descriptive statistics and boxplots. Subsequently, sensitivity, specificity, and predictive accuracy were calculated for each time point. Multi-period ROC curves were plotted, and predictive performance across time periods was compared using statistical methods such as the DeLong test. Dynamic analysis revealed that the model exhibited the highest area under the curve (AUC) and predictive accuracy during the third week of life (days 14-21), identifying this period as the optimal prediction window. Finally, cross-validation and external validation were used to calculate the AUC, confidence interval, and P-value for each time period to ensure the robustness and clinical applicability of the model.

[0097] The steps of building a prediction model include:

[0098] (1) Multidimensional data integration: The collected eosinophil levels, gestational age, and other clinical indicators at each time point are integrated into a multidimensional dataset. Data preprocessing, normalization, and missing value processing algorithms are used to build a high-quality data foundation.

[0099] (2) Feature selection and weight setting:

[0100] Statistical regression, random forest, and LASSO regression algorithms were used for feature selection to clarify the contribution of each indicator to the prediction of BPD. Weightings for each indicator were determined, such as the weight for eosinophils, gestational age, and time series indicators. Cross-validation was used to determine the optimal combination ratio.

[0101] (3) Model training and parameter adjustment: Use logistic regression, support vector machine, or ensemble learning algorithms to build prediction models. During model training, use time series cross-validation techniques to ensure the robustness of the prediction model across time periods. Use model fusion strategies to integrate multiple models to improve overall prediction accuracy.

[0102] (4) Model validation and optimization: Model validation is performed using external datasets to obtain performance indicators such as AUC, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). The model is iteratively optimized based on the validation results until it meets the predetermined clinical application criteria. A prediction threshold is established, and the risk cutoff point is determined using statistical methods to provide a clear reference for clinical decision-making.

[0103] Outstanding predictive accuracy in the third week of life: Time series receiver operating characteristic (ROC) analysis showed that the model's area under the curve (AUC) peaked in the third week of life (days 14-21), with both sensitivity and specificity exceeding those at other time points, demonstrating optimal predictive ability. This result provides a clear time window for early clinical intervention, significantly improving the identification of premature infants at high risk for BPD and the targeting of nursing interventions.

[0104] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for establishing a multidimensional BPD nursing prediction model for premature infants, characterized by: The following steps are involved: Sample collection: Clinical data of several premature infants hospitalized in the NICU were collected, and several premature infants were selected as samples; clinical data of the samples were collected at multiple time points, including blood specimens and imaging data; Inclusion and exclusion: formulate inclusion rules to include samples that meet the inclusion rules; formulate exclusion rules to exclude samples that meet the exclusion rules; Data collection and sample examination: Baseline data were collected using a standardized data collection form; the baseline data included demographic characteristics, perinatal factors, neonatal scores, and resuscitation status; and samples were examined; Grouping and BPD diagnosis: Eosinophil detection was performed, and samples were divided into normal, mildly elevated, moderately elevated, and severely elevated groups based on peak EC levels. BPD was diagnosed using the 2019 Jensen consensus criteria, assessed at 36 weeks of corrected gestational age or at discharge, and graded according to respiratory support method and oxygen requirement. Statistical analysis: denoising, missing value filling and data standardization were performed on the collected data; SPSS 26.0 software was used for statistical analysis. The measurement data obtained from the data collection and sample examination steps, grouping, and BPD diagnosis were integrated into the measurement data for statistical analysis. Statistical analysis was used to detect data anomalies and make corrections, including the following sub-steps: S1. Analysis of Risk Factors for BPD in Premature Infants: Multivariate logistic regression analysis was used to identify independent risk factors for BPD; these independent risk factors included gestational age <28 weeks, Eo, sepsis, and the number of days of invasive ventilation. S2. Statistically analyze the distribution of BPD severity according to peak EC levels and conduct a correlation test between gestational age and BPD risk, demonstrating that BPD risk increases exponentially with decreasing gestational age. S3. Set comprehensive indicator characteristics as area under the curve, sensitivity, specificity, positive predictive value, negative predictive value, and eosinophil level to comprehensively evaluate the predictive efficacy of the model; S4. Retrospective analysis of clinical data of the samples: LASSO regression was used to screen key indicators with predictive value; these key indicators included perinatal indicators, laboratory indicators, respiratory support-related indicators, and complication-related indicators; Model construction: The random forest algorithm is used to model key indicators, comprehensive indicator characteristics, and selected variables, and the model performance is evaluated through 10-fold cross-validation. This specifically includes time series ROC analysis steps and prediction model construction steps; The time series ROC analysis procedure includes the following steps: (1) Calculate the comprehensive index characteristics of each time point and generate the corresponding ROC curve; (2) Use time series data to dynamically evaluate the prediction performance at each time point and draw a multi-period ROC curve change graph; (3) Mining the turning point of the ROC curve, determining the optimal prediction time window, and using statistical methods to compare performance over multiple time periods; (4) Using cross-validation and external validation sets, the consistency of dynamic ROC curves in different samples was tested; (5) Statistically analyze the AUC value, confidence interval, and P-value for each time period to identify the optimal prediction period and provide data support for subsequent clinical decision-making; The steps for constructing the prediction model include: (1) Multidimensional data integration: Integrate the collected eosinophil levels, gestational age, and clinical indicators at each time point into a multidimensional dataset; (2) Use algorithms such as statistical regression, random forest, and LASSO regression for feature selection to clarify the predictive contribution of each indicator to the occurrence of BPD; (3) Determine the weight settings of each indicator, including the weights of eosinophils, gestational age, and time series indicators, and use cross-validation to determine the optimal combination ratio; (4) Construct a prediction model using logistic regression, support vector machine, or ensemble learning algorithms; (5) During the model training process, adopt time series cross-validation techniques to ensure the robustness of the prediction model in each time period; (6) Adopt a model fusion strategy to integrate multiple models to improve the overall prediction accuracy; (7) Apply an external dataset for model validation to obtain performance indicators such as AUC, sensitivity, specificity, positive predictive value, and negative predictive value; (8) Iteratively optimize the model based on the validation results until it meets the predetermined clinical application standards; (9) Establish a prediction threshold and determine the risk cut-off point through statistical methods to provide a clear reference for clinical decision-making.

2. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: The inclusion criteria are preterm infants with a gestational age less than 32 weeks, admitted to the NICU immediately after birth, surviving to a corrected gestational age of 36 weeks or more, and having complete clinical data.

3. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: The exclusion criteria are preterm infants with congenital heart disease, chromosomal abnormalities, or other congenital malformations of important organs; those who died during hospitalization or were transferred to another hospital due to changes in their condition; and cases with a clinical data missing rate exceeding 20%.

4. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: The normal group is defined as EC ≤ 0.4×10^9 / L; the mildly elevated group is defined as 0.4×10^9 / L < EC ≤ 1.0×10^9 / L; the moderately elevated group is defined as 1.0×10^9 / L < EC ≤ 3.0×10^9 / L; the severely elevated group is defined as EC > 3.0×10^9 / L.

5. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: The perinatal indicators include gestational age at birth, birth weight, use of antenatal corticosteroids, and duration of premature rupture of membranes; the laboratory indicators include eosinophil counts and their dynamic change trends, C-reactive protein levels, and platelet counts in the first to third weeks after birth; the respiratory support-related indicators include mechanical ventilation time, inhaled oxygen concentration, and ventilator parameters; the complication-related indicators include early infection status and patent ductus arteriosus status.

6. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: In the steps for constructing the model, in a specific clinical scenario, adjust the prediction strategy and adopt a hierarchical management strategy to adjust the monitoring frequency and intensity of intervention measures according to the predicted risk level.

7. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: In the steps for data collection and sample inspection, the specific inspection of the samples is as follows: Laboratory inspections focus on blood routine results, which are measured using a Sysmex XN-2000 fully automatic hematology analyzer; EC monitoring is performed at least 3 times a week, and the blood sampling time is the early morning fasting period.

8. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: In the statistical analysis step, the measurement data are calculated based on the distribution characteristics using the mean + standard deviation. The data were expressed as mean (Q1), mean (Q3), or median (M(Q1, Q3)). Enumeration data were expressed as number of cases (n(%)). Intergroup comparisons were performed using the t-test, Mann-Whitney U test, or chi-square test. Fisher's exact test was used when the theoretical frequency was less than 5. Spearman correlation analysis was used to evaluate variable correlation, and logistic regression analysis and receiver operating characteristic (ROC) curves were used to assess the predictive value of EC for BPD. All statistical tests were two-sided, and P < 0.05 was considered significant. Statistical graphs were generated using GraphPad Prism 9.0 and R4.1.0 software.

9. The method for establishing a multidimensional BPD nursing prediction model for premature infants according to claim 1, characterized in that: Specific methods for eosinophil detection include: sample collection and processing, collecting blood samples, preferably multiple sampling after birth, on the first day, on the third day, and on the seventh day. Samples must be stored and transported under sterile, low-temperature conditions to ensure the accuracy of the test results; the detection method uses a fully automatic blood analyzer for preliminary screening, combined with flow cytometry to further accurately count eosinophils; uses specific antibodies for immunofluorescence labeling to ensure detection sensitivity and specificity; the results are automatically counted and corrected by the software, and standardized data is output; data verification and quality control: a control group and a repeated detection mechanism are set up at the same time to ensure data consistency; a data quality review system is established to manually review and confirm abnormal values.

Images